Distribution of time-constants for tunneling through a 1D Disordered Chain

Distribution of time-constants for tunneling through a ID Disordered Chain

I

cd

I

o

>

in

en
en
(N

O
On
0^
-(— •

o
o

X

C.J. Bolton-Hcaton*, C.J. Lambert*, Vladimir I. Falko*, V. Prigodin^ and A.J. Epstein f
Department of Physics, Lancaster University, Lancaster LAI 4^B, UK
' Physics Department, The Ohio State University, Columbus, OH 4.3210-1106, USA

(February 7, 2008)

The dynamics of electronic tunneling through a disordered ID chain of finite length is considered.
We calculate distributions of the transmission coefficient T, Wigner delay time and, r^ and the
transport time, Tt = Tt^. The central bodies of these distributions have a power- law form, what
can be understood in terms of the resonant tunneling through localised states.

05.45+b, 03.65Nk, 24.30-v

During the past decade, transport properties of
phase-coherent disordered low-dimensional conductors
attracted much interest. The transmission properties
of complex structures with a large number of internal
degrees of freedom have been intensively studied using
random matrix theory ||l|, and a universal distribution
for transmission coefficients through metallic systems has
been derived . The one-dimensional localization prob-
lem is exactly solvable, and a complete description of the
distribution of transmission coefficients, T and localiza-
tion properties of single-electron wave functions in disor-
dered ID and quasi-lD wires is now available |^,|l^. In
particular, it has been found that the inverse localization
radius, a of single-particle localized states by disorder in
a ID chain has a normal distribution with width inversely
proportional to the chain length, L:

P{a) ^ exp < — L

[a- aoY^

2ao

(1)

Here, Uq is the most probable localization radius which
for a weakly disordered system is given by Uq ~ 41,
where I is the mean free path. This is equivalent to a
log-normal distribution of the transmission coefficients,
T ^ exp(— aL),

P{T) ~ ^ exp .

(ln(l/T) - 2Laof
8Laa

(2)

Recently, random matrix theory was applied to the
problem of the dynamical electric response of mesoscopic
conductors and disordered wires in the localized regime.
Here the problem involves understanding the distribution
of the Wigner delay time, r^ which carries information
about the life-time of carriers in the resonant states re-
sponsible for transmission through a weakly couple quan-
tum dot, or through a disordered wire in the localized
regime. The Wigner delay time is related to the en-
ergy dependence of the phase shift 9(e) of a wave passing
through a disordered wire or a quantum resonator, and
is given by

T4,

d0{e) _ 1 d ^^ GRie,L)
de 2ide ^GA{e,L)'

(3)

This equation also expresses the delay time in terms of
retarded and advanced Green functions, Gr^a{^iL) =
X^i/ i^l{^)^v{L)/ [e — El- ± «0], which links r^ to the den-
sity of states of a system, r^ = tt J„ dxi>{e,x). The
Wigner delay time in zero-dimensional mesoscopic sys-
tems modelled using random matrix theory has been
studied B within the zero-dimensional cr-model ap-
proach, and has been shown to have a universal distribu-
tion. In the present paper, we analyze a related dynami-
cal characteristic of a disordered conductor, namely, the
transport time Tt defined as the delay time weighted by
the transmission coefficient.

Tt = Tta,

(4)

This quantity characterizes the ability of a resonant
state to provide a dynamical response to an external
ac-electric field. Using the Landauer-Buttiker approach,
the imaginary part of the dimensionless ac-conductance
g — G{uj)/'^^ of a single-channel mesoscopic wire pT| , p^ ,

g = T{u) = vlGnie + l^/2, L)GA{e - c^/2, L),

can be represented as

Imgiuj) = nm[l — ilot^ + ...]

-WTt,

SO that the transport time, Tt is a directly measurable
quantity, which can be also interpreted as the dielectric
response function of an almost insulating ID wire, when
ReG — > 0. In the present paper, we report the results of
numerical studies and a qualitative asymptotic analysis
of the transport-time distribution function Pijt) in the
localized regime of ID disordered wires. To anticipate
a little, we find that the distribution of this quantity is
affected by correlations between the value of the Wigner
delay time and the transmission coefficient of resonances
via localized states. Using information about the distri-
bution of the localization radia in Eq. (^ and about the
energetic widths of individual resonances, we show that
the central body of the distribution of Tt, which corre-
sponds to —2/3 < (1/z) ln(rt/r) < 1/3, where z = Lao,
is given by the power-law asymptotic.

Pin)

T-'e'

zl2 I ^
Tt

4/3

(5)

in complete agreement with our numerical simulations.
The tail of short times n, (1/z) ln(Tt/r) < —2/3 decays
in the logarithmically normal way,

, , , ,„ /Tt\-(3/2) + (l/(8z))ln(r/n)

whereas for (1/z) ln(Tf/T) > 1/3,

l+{l/{2z))ln(rt/T)

P(n)^r-ie-/^(^y

(6)

(7)

Below, we show how Eqs. (|^{7|) can be obtained. And
describe the numerical procedure used to determine the
distribution of r^ and Tt . We begin with an analysis of the
Wigner delay time and introduce the numerically-studied
model. In agreement with the result of Ref. ||^, we show,
both analytically and numerically, that the body of the
distribution function P{tii,) is dominated by the inverse-
square-law asymptotic

P(t0) ^t/tI, at T0 > T,

(8)

where r is the mean free path time. These results pro-
vide a check both of both the analytical method and
the numerics and are followed by an the analysis of the
transport-time distribution, which is the central goal of
this paper.

To obtain the distribution of r^, we note that in the ab-
sence of disorder in a ID wire, T = 1 and t^ = Tt ~ L/vp,
vp being the electron Fermi velocity which determines
the ballistic time of flight of an electron through the
chain. For a weakly disordered chain characterized by
a mean free path I = vt or scattering time r <C 1/e,
the transmission T, r^ and Tt are random variables. In
a long wire L ':§> I, where localization is strong, trans-
mission can be viewed as being the result of tunneling
through resonant levels, each characterized by its energy,
eo and decay width, 7 = 71 -I- 72 determined by the elec-
tron escape rates 71.2 into the left and right contacts. In
the exponential localization regime L ^ I, the tunnelling
rates associated with a resonant state peaked at a; < L
(calculated from the left end of the chain) are of order

7i=T-ie-2"i-

and 72 = T t

-2a2{L-x)

(9)

where ai and a2 are two independently fluctuating in-
verse localization radia of the wave function tail on the
left and right hand sides of the resonance. Note that
since the resonance width falls exponentially with the
wire length, whereas the mean level spacing A = l/vL
is only inverse proportional to L, the assumption L ^ I
allows us to distinguish between resonances and to con-
sider each resonant state as a slightly broadened discrete
level.

Therefore, at each value of the energy, the ac-
transmission through the disordered ID chain can be de-
scribed using the Bright- Wigner formula,

T{e)

47172

(e^eor+j^

parametrized by four independently fluctuating parame-
ters: the energetic position cq of the resonance which is
closest to the energy e, the location of the center of mass
of the resonant state, x and two inverse localization ra-
dia, ai and a2. The associated Wigner delay time can
be represented as

1

1

7 1 + [(e - eo)/7]

To analyze the distribution function P{t^), we assume
that the center of the localized state and its energy have
a homogeneous distribution, and that the probability to
find some value of the inverse localization radius ai in
a segment of a wire with the length Xi { xi = x and
X2 — L — x) is equal to

P{a,)

27ra(

■ exp

2a,

■{ai

aa

(10)

It is convenient to use random variables p =
{aixi +a2X2)/z and q = {a\X\ —a2X2)lz, instead of
ai and a2, where z — Lao = L/Al, which can be de-
scribed using the joint distribution function

Pip,q)

1

27ryr

: exp

(p-l?

(g - yp)'^
(1 - y')

(11)

where y = {x/ L) — 2 is uniformly distributed within the
interval [—1,1]. In the same parametrization, one can
represent the Wigner delay time as

T0

exp(0p)

1 + (e/7)^ cosh(zg) '

and the corresponding probability density as the condi-
tional probability integral

nr^) = r

dpdqPip,q) (12)

1 exp(zp) ^ T0 \

{e/-ff cosh{zq) T J ■

To analyze Eq. (O) , we evaluate the integral using the
saddle-point method, and for t^ < re^ find that

Pir^,)

(13)

Note that the length of the chain does not appear in this
result, which indicates that this intermediate asymptotic
of the delay-time distribution, which is related to the dis-
tribution of the spectral width of the resonant states, is
dominated by the electron escape-rate from the resonant

state into the nearest reservoir and for L ^ cxd is exact
for any delay time. However, the finite length L deter-
mines a cut-off, T0 ^ re^ for this universal behavior, and
for T^ > T exp(z) we find

P(r4>)

exp(-^/2) /rV"-^'/^''"'"^^*^^^

T0

in agreement with Melnikov p|.

To illustrate the validity of the above estimates, we
compute the scattering matrix of a series of equal
strength, randomly spaced delta- function scatterers. The
spacings between scatterers possesses a Poisson distribu-
tion, and the system is described by the Hamiltonian

(14)

In the region Xj^i < x < Xj, an eigenstate of energy e
is of the form

V'e(2:) = Aj-e*''^ -I- Bj-e"*''^, Xj-i < X < Xj

where k = \/2me/h and the wave amplitudes on either
side of the scatterer j satisfy

A,+i

i+i _

Bj+i

T,

A,
B,

In this expression T, is the transfer matrix.

T,

1 - iP -iPe-^'^'-"^

(15)

(16)

where (3 = mfj,/h k.

The transfer matrix for a series of A^ scatterers has the
form,

T =

T21 f,

22

(17)

and is given by the product of the transfer matrices of
each individual scatterer

T = TnTn^i • • -Ti,

from which one obtains the scattering matrix

r t'

S =

t r'

via the relation

T[ rrf — 1 rr\ — 1

21J22 J22

T* — 1 rji rrt — 1

11 J12J22

(18)

(19)

For a chain of length L the conductance and the inverse
localization radius are determined by

g — |tp and a{L)

ln\t\^

As an example of Eq. (Q), and to emphasize that we
are simulating a chain of weak scatterers for which a^
is greater than the mean spacing, Fig. 1 shows P{a)
for a chain of scatterers of concentration unity, strength
Uq = i and energy e = A.lir^. These same parameter
values were used in all numerical simulations described
below.

Fig. 2 shows the corresponding plot of In P{t^)t^
for the chain lengths L = 50, 70, 100, 200. These re-
sults were obtained by evaluating a finite difference
T0 — {9{e + S) — 9{e)) /S, and were found to be sta-
ble with respect to any choice of d within the range
10^^ < S < 10"^. For Inr > all curves exhibit a
plateau with a slope that tends to zero with increasing
length, in a good agreement with the analytically esti-
mated asymptotic behavior in Eq. (g).

Having obtained agreement with known results for
P{a) and P{tij,), we now present an analysis of the trans-
port time distribution, P{Tt), which represents the cen-
tral new result of this paper. Using the resonant tun-
neling description of Section II, the transport time for a
particle with a given energy e is given by

Tt = Tts =

471727'

{l + [{e-e„)hfy

(20)

This allows us to parametrize the transport time using
the position x of the resonant state center and two (left
and right) inverse localization radia, ai_2 as in Section
H,

At

Tt

exp(zp)

i + (e/7r

cosh (zq)

The probability to find a given value of Tt can again be
expressed as a conditional probability

P{Tt)^T-'J dy J dpdqPip,q) (21)

4 exp(zp) Tt

+ (e/7)1 ^o'^h (^'^)

and evaluation of the integral in Eq. (K^) using
the saddle-point method yields the result of Eqs.(H-
M). Note that the power-law asymptotic P{Tt) ^

T~^e~^" \-^\ , valid for the finite length wires within

the parametric interval —2/3 < (1/z) ln(Tt/r) < 1/3, for-
mally transforms into the universal central body of the
distribution in the thermodynamic limit i — > 00.

To illustrate the validity of this result. Fig. 3 shows
plots of the function \wP{Tt)T^ versus In Tt for various

lengths. These numerical simulations show that at large
Tt all curves exhibit a plateau with a slope that tends to
zero with increasing length, demonstrating that the tail
in the distribution of Tt , varies as t^ ' .

In summary, we have shown how earlier results for the
Wigner-delay time r^, based on a picture of resonant
transport through localized states, can be extended to
yield the distribution of the transport time Tt = Tt^. In
contrast with the distribution of r^, which exhibits a uni-
versal I/t? tail, the corresponding intermediate asymp-
totic of the distribution of Tt exhibits a universal 1/tj
behavior.

0.0 0.2 0.4 0.6 0.8 1.0

a

FIG. 1. Distribution P{a) versus a for a chain length of
200

[1] B.L. Altshuler and B.D. Simons, in Mesoscopic Quantum

Physics, edited by E. Akkermans, G. Montabaux, J.-L.

Pichard, and J. Zinn- Justin (North-Holland, Amsterdam,

1996).
[2] M. L. Mehta, Random Matrices and Statistical Theory of

Energy Levels (Academic, New York, 1991).
[3] R. A. Jalabert, A. D. Stone, and Y. Alhassid, Phys. Rev.

Lett. 68, 3468 (1992).
[4] Ya.V. Fyodorov and H-J. Sommers, Phys. Rev. Lett, 76

4709 (1996)
[5] P. W. Anderson, D. J. Thouless, E. Abrahams, and D.

S. Fisher, Phys. Rev. B 22, 3519 (1980).
[6] V.I. Melnikov, Sov. Phys. Solid State 23, 444 (1981) [Fiz.

Tverd. Tela 23, 782 (1981)].
[7] R. Landauer and Th. Martin, Rev. Mod. Phys. 66, 217

(1994).
[8] V. Gasparian and M. Pollak, Phys. Rev. B 47, 2038

(1993).
[9] C. Texier and A. Comtet, |cond-mat 9812196i Dec (1998)
[10] C.W.J. Beenakker, Rev. Mod. Phys. 69 731 (1997)
[11] H. M. Pastawski, Phys. Rev. B 46, 4053 (1992).
[12] M. Blittiker, A. Pretre, and H. Thomas, Phys. Rev. Lett.

70, 4114 (1993).
[13] V.A. Gospar, P.A. Mello and M. Biittiker Rev. Lett. 77,

3005 (1996).
[14] R.S. Kohlman, J. Joo, Y.G. Min, A.G. MacDiarmid, and

A.J. Epstein, Phys. Rev. Lett. 77, 2766 (1996).

w 9-

Q.

c

■ length=50

■ length=70

• length=100
length=200

InXip

FIG. 2. Plots of lnP{T^)T^ versus Inr, for various lengths,
, ranging from L = 50 (upper curve) to L = 200 (lower curve.
The size of the ensemble is 10*

^______ .^

1M

"'■ N

-,— ,.

!D

./"A ■

" /

/.

,„„..„, — v^,^_^.: -

/

^ ■ tength=50

>— . Ien9th=70
— -. Iength=100
»--• length=200

-20.0

nxt

FIG. 3. Plots of lnP(rt)rj' versus Inrt, for various
lengths, ranging from L = 50 (upper curve) to L = 200 (lower
curve. The insert shows the central portions of the distribu-
tions. The size of the ensemble is 10*